Process discovery studies ways to construct process models from event logs of historical executions of a system. While discovered models aim to describe the system, process model forecasting aims to construct models that faithfully describe the executions the system will perform in a given period in the future, informing timely system improvements. Existing approaches tackle the problem of process model forecasting by decomposing it into multiple univariate time series forecasting problems. They forecast each directly-follows constraint over a pair of process activities separately and then aggregate these individual forecasts into the resulting process model. In this paper, we propose a deep learning-based approach that leverages multivariate time series forecasting to solve the process model forecasting problem. Our method learns dependencies across all activity constraints simultaneously, generating an integrated forecast of the entire model at once. Through evaluation over industrial event logs, we demonstrate that this approach significantly outperforms existing baselines and statistical multivariate methods in accuracy. Additionally, we introduce a new measure to evaluate the structural correctness of the forecasted models. In the context of information systems engineering, our work addresses the challenge of predicting process models to support future process planning and optimization.